Hugging Face Rebuilds Transformers From the Inside Out for the MoE Era
A massive weight-loading refactor, native expert backends, and expert parallelism make Mixture of Experts models first-class citizens in the transformers library.
Browse all articles in Open Source
A massive weight-loading refactor, native expert backends, and expert parallelism make Mixture of Experts models first-class citizens in the transformers library.
Hugging Face's robotics framework now supports a full humanoid, six new AI policies, and 10x faster training. The gap between lab and garage is shrinking fast.
Georgi Gerganov's llama.cpp and GGML join Hugging Face to build the future of local AI inference. Here's why this merger changes everything.
H Company and NVIDIA release Holotron-12B, an open-source computer-use agent with 2x throughput gains on a single GPU.
Mistral Small 4 unifies reasoning, multimodal, and coding capabilities into a single 119B-parameter open-source model under Apache 2.0.